Quality versus accuracy: result of a reanalysis of protein-binding microarrays from the DREAM5 challenge by using BayesPI2 including dinucleotide interdependence

Table 2 Correlations between the estimated TF-binding affinities based on inferred PBEM from in vitro data and the measured TF tag counts from in vivo ChIP-seq experiment

ChIP-Seq	Called peaks	Rank	Paired PBMs agreement	R (top 500)	R (top 1000)	R (top 2000)	R (all peaks)
TF_31 (Zfx)	10338	8	Good	0.38978	0.35943	0.33566	0.25502
TF_44 (Gata4)	16979	22	Good	0.29675	0.28458	0.2645	0.15614
TF_23 (Tbx20)	4012	33	Good	0.077781	0.062428	0.043291	0.043338
TF_25 (Tbx5)	56352	41	Bad	0.062865	0.090766	0.073079	0.053016
TF_40 (Esrrb)	21647	59	Bad	0.16334	0.12283	0.10351	0.064026

In the table, the first column describes ID of mouse TFs in DREAM5 challenges and the TF names to the ChIP-seq experiments; Called peaks are the number of called peaks in the ChIP-seq data; Rank represents rank order of TFs that were sorted in decreasing order of the final performance score across all tested algorithms in Figure 2 of original publication [1]; paired PBMs agreement indicates the agreement between training and testing PBMs; R is square root of R-square statistic from a linear regression model, where the relationship between the predicted TF-binding affinities and the observed ChIP-seq tag counts is investigated.

ISSN: 1471-2105